An AI-native code review knowledge base with structured rules that catch what linters miss. Open source, MIT licensed.
BeforeMerge is an AI-native code review knowledge base — a structured collection of rules organized into skills that teach AI reviewers (and humans) what to look for in pull requests.
Unlike traditional linters that focus on syntax and formatting, BeforeMerge rules encode architectural decisions, security best practices, and framework-specific patterns that are typically only caught by senior engineers during manual review.
Every team has unwritten rules. Things like:
getServerSideProps for static content in Next.js"These rules live in engineers' heads. They get shared in PR comments, forgotten in onboarding docs, and inconsistently enforced. Senior engineers spend hours repeating the same feedback.
We wanted to create a single source of truth — a knowledge base that's structured enough for AI to consume, but clear enough for any developer to learn from.
A skill is a collection of related rules for a specific technology. For example, the supabase-nextjs skill contains rules about Row Level Security, edge function patterns, auth handling, and more.
Each rule is a structured document with:
Rules are designed to work with AI code review tools. Pipe them into your review workflow, and your AI reviewer gains the collective knowledge of experienced engineers.
The entire knowledge base is MIT licensed. We believe code review knowledge should be shared, not locked behind proprietary walls.
You can:
We're building an automated PR review service that applies these rules to your pull requests on every push. Join the waitlist to get early access.
In the meantime, explore the rules, star the repo, and let us know what skills you'd like to see next.